ChatGPT and similar tools have caused quite a stir in recent months. They can be an enormous help, not least when it comes to content creation. At the same time, you should be aware of the limitations of these tools and understand how to achieve particularly good results.

In this article, I explain why the new generation of AI offerings is so much better. It is the start of a series on the topic. The topic of "artificial intelligence" has been experiencing phases of exuberance and depression for decades now.

Elon Musk's predictions regarding autonomous vehicles, for example, are well known: the field made rapid progress for a while. Computer-controlled cabs and buses seemed to be within reach. However, the curve of improvement soon flattened out and forecasts in this area have since become much more cautious. In this respect, I have become accustomed to a healthy skepticism when it comes to such hype topics. 

However, this quickly flew out the window with ChatGPT. When I was able to try it out for the first time, I was as amazed as I was thrilled: this "chatbot" finally worked the way its many predecessors had only promised. It was almost scary.

ChatGPT responds to questions in an amazingly human way. And it seems to have an answer to every question - or several. It dynamically adjusts the type, length and complexity of the output based on my input. It understands the context of the conversation and can draw on topics and facts that have been raised previously. It processes long and complex inputs with little delay. And it also understands and responds in German.

Amazingly powerful text generator

It quickly became clear that ChatGPT is not only a powerful AI assistant, but also an AI text generator.

My previous attempts with tools of this type had always been big disappointments. The products could never keep up with the marketing promises. They were perhaps suitable for stimulating ideas. You could extract a few set pieces. But the texts were rarely useful as a whole.

The situation is different with ChatGPT and its variants and competitors: Used correctly, they can not only provide ideas, but also a comprehensive concept and at least a good first draft.

As someone who spends a large part of their livelihood writing and editing text, I hate to say this: what ChatGPT delivers is often on the same level as what I've seen from human writers. Sometimes it's even better.

However, the limits and special quirks of these new AI tools also become clear after intensive testing. More on this below.

How was this progress made possible?

But how was this sudden leap in quality even possible? Three points are usually mentioned:

1. Training data: Today's AI models learn using existing content (text, images, audio, code, etc.). The amount of data already available in digital form has increased rapidly, which helps training enormously.
2. Computing power: Specialized computers and components have significantly accelerated the training processes and enable larger and more complex models.
3. algorithms: Last but not least, there have been important advances under the hood. The "Transformer" method, for example, is one of the main reasons why AI can understand and generate texts so much better today than it could a few years ago.

It is also impressive to look at the number of "parameters" of language models in recent years. More parameters enable the model to encode more knowledge and handle more complex tasks:

• 2019, GPT-2: 1.5 billion parameters
• 2020, GPT-3: 175 billion parameters
• 2022, PaLM: 540 billion parameters
• 2022, GPT-4: around 1 trillion parameters
• 2022. Claude: around 10 trillion parameters

These figures are fascinating, but more complex models are not always automatically superior or the better choice. There is currently a trend towards training models more specifically for certain tasks and adjusting them accordingly. It is therefore to be expected that, in addition to general tools, there will be more and more offerings that are intended for a clearly defined purpose. 

Models that are additionally trained with individual data are also exciting in this context: For example, companies can feed all of their documents into such a system in order to receive dynamic answers.

Another interesting measurement value that has recently moved more into the spotlight is the context length. The greater this value, the more content the tool can include in a conversation. More context therefore helps the AI to conduct longer chats, but also to process larger inputs. 

Depending on the use case, this can make a considerable difference when a service such as Anthropics Claude processes and analyzes entire books within seconds.

The context length is measured in "tokens", whereby one token corresponds approximately to one word. Some examples:

• GPT-2: 1,024 tokens
• GPT-3: 2,048 tokens (in a new version up to 16,000 tokens)
• PaLM: 65,536 tokens
• GPT-4: up to 32,000 tokens
• Claude: probably around 100,000 tokens

A longer context requires correspondingly more computing power and storage space. It is therefore a technical challenge to further increase these values.

Three options for using such tools

Anyone who wants to use such tools currently has three main options:

1. In the cloud. ChatGPT, Claude, but also image generators such as MidJourney or Stable Diffusion can be used as software-as-a-service. This means that your own data is processed on the provider's servers. Depending on the type of information, this can be problematic. At the same time, as a user, you have to be satisfied with the interface and the options offered. Companies such as OpenAI, Microsoft, Google and Anthropic have specialized, particularly powerful servers for this purpose. 
2. Via an API. OpenAI in particular actively offers its interfaces. Not all AI models are immediately available to everyone. Nevertheless, you can either implement your own applications or use third-party apps. Data processing continues to take place on the servers of the AI companies. Where and how the offer can be used is individually customizable in this case.
3. On your own computer or server. Not only specialized computers have become more powerful, but also generally available laptops, tablets and even smartphones. With modern and appropriately equipped devices, this can be enough to use tools such as AI assistants directly on your own computer. They are not as powerful as the high-end applications in the cloud. But that is not always necessary. Instead, your own data remains on your computer. In addition, the software and model can be selected entirely according to your own needs. One example is LM Studio for Windows and Macs, which allows you to use language models such as Meta's Llama family on your own PC.

In addition, there is currently another trend that I believe will become even more prevalent: AI assistants that are integrated into other offerings. Examples include "Copilot" in Microsoft 365, Adobe's "Firefly", Bing Chat and Google's experimental, AI-supported "Search Generative Experience" (SGE).

Limits of AI tools

In further articles in this series, I will show you in more detail how I personally use such services to research topics, generate ideas and concepts and create texts and images. 

Despite all the enthusiasm for the opportunities and possibilities offered by these new little helpers, they have limitations that you should be aware of, and there is justified criticism.

A service like ChatGPT, for example, has learned to give a linguistically correct and meaningful-sounding answer. That is the focus. The validity of the facts and figures mentioned, on the other hand, is not. They may be true or they may be fictitious. You should therefore not accept the statements without checking them.

For some tasks, these tools are also completely overwhelmed. For example, they are often not good at dealing with numbers and calculations.

The providers are trying to counteract this. On the one hand, the AI assistants are to be trained to be more honest. If they don't know something exactly, they should make this clear. On the other hand, OpenAI has added plugins as an option: This allows ChatGPT to access specialized tools and information sources for certain topics and tasks. Bing Chat is another example: it provides links to the sources of its answers and makes it clear if it was unable to find a piece of information.

Moreover, the knowledge of an AI assistant such as ChatGPT or Claude often only extends up to a certain date. Everything that has happened since then is unknown. The training process of such an AI is so complex and lengthy that new information cannot simply be added. You have to be aware of this for some topics.

Another problem is that an AI can spread prejudices and false information that it has found in its training data, thereby reinforcing them. After all, the AI does not understand what it is doing there. Nor does it normally check or research.

What I also miss from time to time with AI assistants in everyday life is that they don't get to know me and they don't learn from previous conversations. As described above, there is a certain context length per chat. But the context always ends with the current chat. If I start a new conversation, the AI assistant knows nothing about previous interactions. My hope is that these services will become even more personalized in the future. SHO.AI promises that, for example.

Criticism of the AI tools

A fundamental criticism of tools such as ChatGPT for texts or Stable Diffusion for images is the training material. As already described, this data is indispensable for the learning process. However, the authors were often not asked whether they wanted to make their work available for this purpose or not. For example, the fact that AI image generators can imitate artists' styles caused a stir. Is this an automated copyright infringement? Or is it comparable to human works, which can ultimately also be inspired and influenced by the works of others? These are exciting questions that will be with us for years to come. 

The debate about this is sometimes heated. No wonder: some artists see themselves as involuntary stooges for an AI that could make them superfluous in return. And the companies earn money with a product that has used their work for free.

OpenAI now offers an option to at least block the content of your own website for such training purposes in the future.

This also raises the question of whether the results of such tools may be used at all. I spoke to lawyer Dr. Carsten Ulbricht about this. As is so often the case, the question cannot be answered with a clear yes or no.

Last but not least, the question of whether works enjoy copyright protection if they originate from an AI and who is considered the author in this case is completely open. In the opinion of some, the benchmark here is how much work the AI has done and how much the human has done. 

Conclusion on content and AI

The AI world has experienced a boom and hype in recent months. As I hope I have been able to show in this article, the enthusiasm is not completely out of thin air. The progress is clearly noticeable. The tools can be used for everyday tasks and can be a great help.

With all that said, they are not perfect, they make mistakes, they react unexpectedly or they completely fail at a task (and perhaps even deny it). Furthermore, there is justified criticism of how these tools work and how they have acquired their skills.

With these points in mind, in the next part of the series I will show you how I use various AI tools for creativity and productivity.